1. Working in different contexts: RStudio Projects
2. Dynamic document generation: RMarkdown
3. Version control: Git + GitHub
4. Package management: renv
5. Containerization: Docker
6. Where to start?

• Peikert, A., & Brandmaier, A. M. (2020). A Reproducible Data Analysis Workflow with R Markdown, Git, Make, and Docker. https://doi.org/10.31234/osf.io/8xzqy

• Andreas Brandmaier: “A Reproducible Data Analysis Workflow with R Markdown, Git, Make, and Docker”, talk at the Online Symposium on Reproducing Analyses in Biological Psychology, November 2020 (https://www.youtube.com/watch?v=nTVcMDVlyOI)
• Richard McElreath: “Science as Amateur Software Development”, talk at the Max Planck IT Community, August 2020 (https://www.youtube.com/watch?v=zwRdO9_GGhY)
• Russ Poldrack: “Toward a Culture of Computational Reproducibility”, talk at the Foundations of Biomedical Data Science, December 2020 (https://www.youtube.com/watch?v=XjW3t-qXAiE)

• What it does:
  • Allows to work in multiple different contexts (projects), e.g. one for each experiment
  • Each project is own working directory, workspace, history, and source documents
  • Each project is associated with a folder on your computer (= working directory)
• Why it helps:
  • Have a separate, shareable working environment for each experiment
  • Keep all the files associated with a project together — data, scripts, results, figures
  • Work on multiple projects at once, each associated with its packages (and package versions), loaded data, etc.
  • Use only relative paths
  • Necessary basis for version control

• In RStudio: File > New Project > …

• Relative paths: path separator characters vary across systems & anchor points differ depending on contexts
  • Use the here-package (Müller, 2020) to define relative paths within the project: read.csv(here::here("data", "file_I_want.csv"))

• What it does:
  • Creates dynamic documents with embedded chunks of code (R, python, Julia, stan, …), computed results , written text etc. (= LaTeX)
  • Markdown-files can be exported to documents (docx, rtf), presentations, pdfs, websites (html), … e.g using the knitr (Xie, 2015, 2020) and tinytex (Xie, 2015, 2020; for pdfs)
  • R code is dynamically rendered, and can be given in separate chunks (’’‘{r}’’‘) or inline (’ r … ’)
• Why it helps:
  • Simple language (\(\neq\) LaTeX)
  • Integrates directly with statistical software (R Studio)
  • Saves code and output in one file
  • Reduces copy&paste errors: reported results consistent with actual results

• Installation: install.packages("rmarkdown")

• Installation: install.packages("rmarkdown")
• Open a markdown file: File > New File > R Markdown

• Installation: install.packages("rmarkdown")
• Open a markdown file: File > New File > R Markdown

• You don’t have RStudio installed: install Pandoc (http://pandoc.org) before installing markdown ()
• Lengthy R code chunks: Install knitr-package (Xie, 2014, 2015, 2020) to customize chunks and knitting process
  • {r cache=TRUE,message=FALSE,warning=FALSE,results="hide", error = TRUE}
  • or use opts_chunk$set()-function
• Knit to pdf: You need a LaTeX-installation
  • TinyTeX (Xie, 2010) is a light-weight, cross-platform distribution (install.packages("tinytex"); tinytex::install_tinytex()))
  • Separate code chunks by a blank line

• What it does:
  • Tracks changes to files (data and code) over time: Sequence of “snapshots” (commits) wit unique identifiers (hash code) and describition (commit message)
  • Allows to “go back in time”: Recall older versions or to revert the entire project
  • Changes between commits can be compared
  • Organized in repositories: Collection of all snapshots
  • GitHub: Popular server for sharing materials (privately or publicly) and collaborating via git (also: GitLab and others)

• Why it helps:
  • Keep things organized and track changes
  • Clean up code
  • Language agnostic
  • (Remote) backup
  • Work together (even simultaneously and parallely: branches, merges, pull requests)
  • Web interface to track issues
  • Easily connected e.g. to osf.io

• …

• Github: No long-term guarantee for availability of service (is commercial)
  • Mirror snapshots on hu servers/osf/zenodo/FigShare/…

• What it does:
  • Creates a project-specific library of packages in the project folder
  • Overwrites install.packages() to install packages in this local library
  • Keeps track of package versions in the renv.lock file

• Why it helps:
  • Keeps package versions untouched by other projects
  • Allows you to revert to the previous state when an update has broken your analysis
  • Makes it easier to share package versions with your collaborators (e.g., via GitHub)

1. Install renv just like any other R package via install.packages(renv)
2. Initialize your project library via renv::init()
   (Instead, you can also select “Use renv with this project” during project creation)
3. After successfully installing or updating packages, use renv::snapshot()
4. If you want to revert to previous state (e.g., if an update caused problems), use renv::restore()

• Initialize renv for your sleepstudy project using renv::init()
• From the Files pane in RStudio, take a look at the renv.lock file
• Install a new package:

install.packages("cowsay")

• Acutally use the package in one of your scripts (or .Rmd files):

cowsay::say("Hello world", "cow")

• Write this change to the lockfile using renv::snapshot()
• Commit and push your changes to GitHub

Restoring someone else’s package versions:

1. Clone or pull the repository from GitHub
2. Open the the RStudio project (e.g. via the projectname.Rproj file)
3. Use renv::restore() to install the package versions from the renv.lock file

• There may be some (inconsequential) warnings when switching between Mac and Windows
• At least on Windows, you need to have Rtools installed when installing packages that are not on CRAN (https://cran.r-project.org/bin/windows/Rtools/)
• Installing and loading packages may take a while, especially if your project lives on a network drive
  (such as N:/)

• What it does:
  • Creates a small, linux-based virtual machine on your computer
  • Makes it possible to run your scripts (or render your .Rmd files) on this virtual system
  • The recipe to build this system is stored in a Dockerfile that can be shared via GitHub

• Why it helps:
  • Prevents differences between operating systems, R versions, region and language settings etc.
  • Ensures long-term reproducibility
  • Provides a starting point for cloud-based and high perfomance computing (HPC)
  • Pre-packaged Docker images are available for lots of different languages (R, Python, MATLAB, LaTeX etc.)

• Install Docker from https://docs.docker.com/installation/
• From the terminal, start a R/RStudio container and mount your local project folder:

docker run -d  -e PASSWORD=1234 -p 8787:8787 -v /path/to/your/project:/home/rstudio/ rocker/rstudio

• You can then access RStudio (running in the container) by opening http://localhost:8787 in your web browser (username: rstudio, password: 1234)
• You can also build your own container by:
  • Choosing a base image from https://hub.docker.com/u/rocker (including the tidyverse, LaTeX etc.)
  • Creating a Dockerfile in your project directy, specyfing additional steps to execute when building the container, e.g., install.packages("renv"); renv::restore()

• For detailled instructions, see Boettiger & Eddeulbuettel (2017) and
  Peikert & Brandmaier (2020)

• Example for a Dockerfile:

# This as a text file stored with the name "Dockerfile" in your project directory.

# Base image from Docker Hub, including R, RStudio, the tidyverse, and LaTeX
FROM rocker/verse:4.0.2

# Set working directory within the container
WORDIR /home/rstudio

# Install renv
RUN R -e "remotes::install_version('renv', version = '0.12.0', repos = 'http://cran.us.r-project.org')"

# Copy the lock file
COPY renv.lock renv.lock

# Install package versions stored in the lockfile
RUN R -e "renv::consent(provided = TRUE)"
RUN R -e "renv::restore(prompt = FALSE)"

• Some additional tools based on Docker:
  • With binder (https://mybinder.org) and Code Ocean (https://codeocean.com), you can run your analysis in the cloud; they will even create the Dockerfile for you if you don’t have your own one
  • Singularity (https://sylabs.io) is a fully compatible, open source clone of Docker which you can use on systems where you don’t have root access (e.g., on high performance clusters)

• This wealth of tools can seem overwhelming
  • Adopting even one or two of them can help making your code more reproducbile
• An RStudio project and renv are easy to set up even for existing projects
  • And will help a lot to make sure that you can still run your code at re-submission
• Version control is best tried out with a new (real or toybox) project
  • Create an empty repository on GitHub and use it to create your RStudio project
• Once you’ve made it this far, full computational reproduciblity (by containerizing your project) is just one more step away